Tower Servers vs. Rack Servers vs. Blade Servers

Continuing my hardware series let me try to put the basic differences between above three.

Tower Servers are the normal boxes (in appearance) you would have seen in Visio Diagrams. Of course they are very powerful & have bundled software tools to manage them. The problem with Tower Severs is the space they occupy, management personnel they require, and cost of operating them (power, network, etc.).

Rack Servers are servers mounted inside a Rack (something like we normally use to manage our letters, office files, etc.) Major Racks available out there adhere to an IEEE standard and are measured in rack units or “U’s” (each U is 19” wide and 1.75” tall). So a rack server size is typically in multiplication of these “U’s”. Motivation here is to scale vertically than horizontally with more compact physical servers. In addition to this, there are many other electronic devices which adhere to this IEEE standard for instance – Rack Consoles, SAN devices, Power Backup devices, etc. Advantage being that you can fix them into rack as well along with your servers. Not to mention that the hardware vendors (Dell, HP, IBM, etc.) provide additional software tools that help you effectively manage these servers and in some cases the supported devices also.

Blade Servers are an additional level of innovation on top of Rack Servers. Blade Servers are typically placed inside a blade enclosure, and together they form a blade system. A Blade system normally meets the IEEE standard of Rack Units, which means that the entire Blade system can be placed inside the rack along with other electronic equipments. The benefits of blade enclosure includes hot plugging (normally blade servers have a handle attached to them, for transferring them in and out of the blade enclosure – it’s an easy way of identifying them) and stripped modular design (e.g. shared network ports, power connections, switches, etc.). For instance the hardware we ordered allows us to pack 16 blade servers inside a 10U space. I remember meeting an Oracle Consultant few months back where he was touting about a server with no disk. Such scenarios are possible (and are cheaper) by coupling a blade system and SAN storage. All these boils down to further space reduction, cost savings (power, administration staff) & easy management. Bundle this with Virtualization and you have a very powerful infrastructure at your disposal.

So any guesses as to what the next generation of servers would be ?

RAID 0 vs. RAID 1 vs. RAID 4 vs. RAID 5 vs. RAID 10

My company had ordered some hardware recently. As usual, the buying process brought in lot of insights which I thought of sharing with you guys. I will cover them all in parts, starting with basics. First let’s see a fundamental concept called RAID which becomes critical when you start planning your storage.

RAID (Redundant Array of Independent Disks) – This technology is mainly used for hard disks to provide protection against hard drive failures and also to improvise performance (of course you can select either of failure protection / performance). RAID also gives an implicit way to scale your disks (after all it’s an Array). RAID builds on 2 primary concepts – Striping (storing data across multiple disks sectors of multiple drives so that disk head can read more data in a single move) and Mirroring(replication of drives). RAID requires minimum of 2 disks & can have maximum of 32 disks. It’s also important to note here that RAID controllers typically abstract out the striping & mirroring details giving end user a feel of only a single disk available at disposal. Below are some of the popular RAID configurations out there:

RAID 0 – Striping
Pros: Better performance – data replicated across drives, no storage overhead as drives are utilized 100%
Cons: Possibility of losing entire data on failure of a single disk
Minimum Disks Required – 2

RAID 1 – Mirroring
Pros – Guard against disk failure as data is replicated across disk drives
Cons – Replication creates storage overhead as the same data is copied across drives
Minimum Disks Required – 2

RAID 4 – A Mirroring variation. It actually uses the concept of ‘Parity’ (based on XOR operation – Sum of all bits across drives – if sum is even parity set to 0 else parity set to 1) to retrive data in case of a disk failure, which reduces storage overhead – an issue with RAID 1.
Pros – Reduced storage overhead (actually we need only a single disk here to store parity). E.g. if you have 3 disks, parity can be stored on 3rd. So your overhead is only 33% in terms of storage.
Cons – Still suffers from a performance prespective
Minimum Disks Required – 3 (anyways 2 doesn’t make sense and the more no. of disks you have lesser would be your storage overhead)

RAID 5 – Striped Data & Striped Parity – most widely used. Taking RAID 4 to next level by striping the parity (parity gets stored across disks instead of getting stored on single disk) and also striping data (that anyways doesn’t matter as far as parity is concerned)
Pros – Good Performance, Good failure protection
Cons – Not as good when your requirement is only performance or only failure protection (parity doesn’t come for free).
Minimum Disks Required – 3

RAID 10 (1 + 0) – Combination of Stripe & Mirror with no parity. You stripe the data across drives & mirror the entire set of drives.
Pros: Best in terms of performance & guards against potential failures.
Cons: Costly in terms of storage overhead
Minimum Disks Required – 4 (and for > 4 you must have even number of disks)

There are few other RAID levels which I would leave for the reader to digg into, as they are comparatively not that widely used. I will continue exploring hardware aspects further in next thread.